---
dataset_info:
  features:
  - name: scene
    dtype: int32
  - name: altitude
    dtype: int32
  - name: frame_id
    dtype: int32
  - name: image
    dtype: image
  - name: depth_map
    dtype: array3d
  - name: voxel_grid
    dtype:
      label:
        dtype: uint8
        shape:
        - 192
        - 128
        - 128
      invalid:
        dtype: bool
        shape:
        - 192
        - 128
        - 128
      occluded:
        dtype: bool
        shape:
        - 192
        - 128
        - 128
      surface:
        dtype: bool
        shape:
        - 192
        - 128
        - 128
  - name: calibration
    dtype:
      K:
        dtype: float32
        shape:
        - 3
        - 3
language:
- en
license: cc-by-nc-sa-4.0
tags:
- 3d scene understanding
- 3d-scene-completion
- aerial perception
- autonomous flying
- dataset
- benchmark
task_categories:
- image-to-3d
task_ids:
- semantic-segmentation
---

# OccuFly Dataset

Following its acceptance as a **CVPR 2026 Oral**, we release OccuFly: the first real-world, large-scale camera-based benchmark for Semantic Scene Completion (SSC) and Metric Monocular Depth Estimation from the aerial perspective.

🌐 **Project Page**: [markus-42.github.io/publications/2026/occufly/](https://markus-42.github.io/publications/2026/occufly/)  
🤗 **Aerial DepthAnything2**: [huggingface.co/markus-42/OccuFly-DepthAnythingV2](https://huggingface.co/markus-42/OccuFly-DepthAnythingV2)  
📄 **Paper**: [arXiv:2512.20770 ](https://arxiv.org/abs/2512.20770)  
📚 **Full Documentation GitHub**: See [github.com/markus-42/occufly](https://github.com/markus-42/occufly)  

---

## Quick Start

### 1. Install Dependencies

```bash
pip install huggingface-hub tqdm numpy Pillow
```

### 2. Download Dataset

Use the download script located in this repository:

```bash
# Download all scenes
python download_occufly.py

# Download specific split
python download_occufly.py --split train
python download_occufly.py --split validation
python download_occufly.py --split test

# Download specific scenes (1-9)
python download_occufly.py --scenes 1 2 3

# Include predicted depth maps
python download_occufly.py --include_depth_predictions
python download_occufly.py --split train --include_depth_predictions

# Download only predicted depth maps
python download_occufly.py --only_depth_predictions

# Custom output directory
python download_occufly.py --output ./my_data
```

## Dataset Splits

- **Train**: Scenes 1-5
- **Validation**: Scenes 6-7  
- **Test**: Scenes 8-9

Each scene has 3 altitudes: 30m, 40m, 50m

## Full Documentation

For detailed documentation, visit [github.com/markus-42/occufly](https://github.com/markus-42/occufly).

## Citation

```bibtex

@inproceedings{gross2025occufly,
    title={{OccuFly: A 3D Vision Benchmark for Semantic Scene Completion from the Aerial Perspective}}, 
    author={Markus Gross and Sai B. Matha and Aya Fahmy and Rui Song and Daniel Cremers and Henri Meess},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2026},
}
```

## License

This dataset is licensed under the **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International** (CC BY-NC-SA 4.0) license.

- **You can**: Share and adapt the material for non-commercial purposes
- **You must**: Give credit, indicate changes, and use the same license for derivative works
- **You cannot**: Use the material for commercial purposes without permission

For full license details, see [creativecommons.org/licenses/by-nc-sa/4.0/](https://creativecommons.org/licenses/by-nc-sa/4.0/)